Parent Directory | Revision Log
Revision 702 - (view) (download)
1 : | Isibaar | 616 | // 30.10.2002 corrected qpel chroma rounding |
2 : | Isibaar | 581 | // 04.10.2002 added qpel support to MBMotionCompensation |
3 : | edgomez | 195 | // 01.05.2002 updated MBMotionCompensationBVOP |
4 : | // 14.04.2002 bframe compensation | ||
5 : | suxen_drol | 118 | |
6 : | Isibaar | 3 | #include "../encoder.h" |
7 : | #include "../utils/mbfunctions.h" | ||
8 : | #include "../image/interpolate8x8.h" | ||
9 : | suxen_drol | 701 | #include "../image/reduced.h" |
10 : | Isibaar | 3 | #include "../utils/timer.h" |
11 : | suxen_drol | 118 | #include "motion.h" |
12 : | Isibaar | 3 | |
13 : | #define ABS(X) (((X)>0)?(X):-(X)) | ||
14 : | #define SIGN(X) (((X)>0)?1:-1) | ||
15 : | |||
16 : | suxen_drol | 701 | |
17 : | edgomez | 195 | static __inline void |
18 : | Isibaar | 588 | compensate16x16_interpolate(int16_t * const dct_codes, |
19 : | uint8_t * const cur, | ||
20 : | const uint8_t * const ref, | ||
21 : | const uint8_t * const refh, | ||
22 : | suxen_drol | 701 | uint8_t * const refv, |
23 : | Isibaar | 588 | const uint8_t * const refhv, |
24 : | suxen_drol | 701 | uint32_t x, |
25 : | uint32_t y, | ||
26 : | Isibaar | 588 | const int32_t dx, |
27 : | const int32_t dy, | ||
28 : | const uint32_t stride, | ||
29 : | const uint32_t quarterpel, | ||
30 : | suxen_drol | 701 | const int reduced_resolution, |
31 : | const uint32_t rounding) | ||
32 : | Isibaar | 588 | { |
33 : | suxen_drol | 701 | |
34 : | if (reduced_resolution) | ||
35 : | { | ||
36 : | suxen_drol | 702 | const uint8_t * reference; |
37 : | x*=2; y*=2; | ||
38 : | Isibaar | 588 | |
39 : | suxen_drol | 702 | reference = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
40 : | |||
41 : | filter_18x18_to_8x8(dct_codes, cur+y*stride + x, stride); | ||
42 : | filter_diff_18x18_to_8x8(dct_codes, reference, stride); | ||
43 : | Isibaar | 588 | |
44 : | suxen_drol | 702 | filter_18x18_to_8x8(dct_codes+64, cur+y*stride + x + 16, stride); |
45 : | filter_diff_18x18_to_8x8(dct_codes+64, reference + 16, stride); | ||
46 : | Isibaar | 588 | |
47 : | suxen_drol | 702 | filter_18x18_to_8x8(dct_codes+128, cur+(y+16)*stride + x, stride); |
48 : | filter_diff_18x18_to_8x8(dct_codes+128, reference + 16*stride, stride); | ||
49 : | suxen_drol | 701 | |
50 : | suxen_drol | 702 | filter_18x18_to_8x8(dct_codes+192, cur+(y+16)*stride + x + 16, stride); |
51 : | filter_diff_18x18_to_8x8(dct_codes+192, reference + 16*stride + 16, stride); | ||
52 : | suxen_drol | 701 | |
53 : | suxen_drol | 702 | transfer32x32_copy(cur + y*stride + x, reference, stride); |
54 : | suxen_drol | 701 | |
55 : | }else{ | ||
56 : | if(quarterpel) { | ||
57 : | interpolate16x16_quarterpel((uint8_t *) refv, (uint8_t *) ref, (uint8_t *) refh, | ||
58 : | (uint8_t *) refh + 64, (uint8_t *) refhv, x, y, dx, dy, stride, rounding); | ||
59 : | |||
60 : | transfer_8to16sub(dct_codes, cur + y*stride + x, | ||
61 : | refv + y*stride + x, stride); | ||
62 : | transfer_8to16sub(dct_codes+64, cur + y*stride + x + 8, | ||
63 : | refv + y*stride + x + 8, stride); | ||
64 : | transfer_8to16sub(dct_codes+128, cur + y*stride + x + 8*stride, | ||
65 : | refv + y*stride + x + 8*stride, stride); | ||
66 : | transfer_8to16sub(dct_codes+192, cur + y*stride + x + 8*stride + 8, | ||
67 : | refv + y*stride + x + 8*stride+8, stride); | ||
68 : | |||
69 : | } | ||
70 : | else | ||
71 : | { | ||
72 : | suxen_drol | 702 | const uint8_t * reference = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
73 : | suxen_drol | 701 | |
74 : | transfer_8to16sub(dct_codes, cur + y * stride + x, | ||
75 : | reference, stride); | ||
76 : | transfer_8to16sub(dct_codes+64, cur + y * stride + x + 8, | ||
77 : | reference + 8, stride); | ||
78 : | transfer_8to16sub(dct_codes+128, cur + y * stride + x + 8*stride, | ||
79 : | reference + 8*stride, stride); | ||
80 : | transfer_8to16sub(dct_codes+192, cur + y * stride + x + 8*stride+8, | ||
81 : | reference + 8*stride + 8, stride); | ||
82 : | } | ||
83 : | Isibaar | 588 | } |
84 : | } | ||
85 : | |||
86 : | static __inline void | ||
87 : | Isibaar | 581 | compensate8x8_interpolate(int16_t * const dct_codes, |
88 : | uint8_t * const cur, | ||
89 : | const uint8_t * const ref, | ||
90 : | const uint8_t * const refh, | ||
91 : | const uint8_t * const refv, | ||
92 : | const uint8_t * const refhv, | ||
93 : | suxen_drol | 702 | uint32_t x, |
94 : | uint32_t y, | ||
95 : | Isibaar | 581 | const int32_t dx, |
96 : | const int32_t dy, | ||
97 : | const uint32_t stride, | ||
98 : | const uint32_t quarterpel, | ||
99 : | suxen_drol | 701 | const int reduced_resolution, |
100 : | Isibaar | 581 | const uint32_t rounding) |
101 : | Isibaar | 3 | { |
102 : | suxen_drol | 701 | if (reduced_resolution) |
103 : | edgomez | 195 | { |
104 : | suxen_drol | 702 | const uint8_t * reference; |
105 : | x*=2; y*=2; | ||
106 : | |||
107 : | reference = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); | ||
108 : | |||
109 : | filter_18x18_to_8x8(dct_codes, cur+y*stride + x, stride); | ||
110 : | filter_diff_18x18_to_8x8(dct_codes, reference, stride); | ||
111 : | |||
112 : | transfer16x16_copy(cur + y*stride + x, reference, stride); | ||
113 : | |||
114 : | suxen_drol | 701 | } else { |
115 : | |||
116 : | if(quarterpel) { | ||
117 : | interpolate8x8_quarterpel((uint8_t *) refv, (uint8_t *) ref, (uint8_t *) refh, | ||
118 : | (uint8_t *) refh + 64, (uint8_t *) refhv, x, y, dx, dy, stride, rounding); | ||
119 : | Isibaar | 581 | |
120 : | suxen_drol | 701 | transfer_8to16sub(dct_codes, cur + y*stride + x, |
121 : | refv + y*stride + x, stride); | ||
122 : | } | ||
123 : | else | ||
124 : | Isibaar | 581 | { |
125 : | suxen_drol | 702 | const uint8_t * reference = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
126 : | suxen_drol | 701 | |
127 : | transfer_8to16sub(dct_codes, cur + y * stride + x, | ||
128 : | reference, stride); | ||
129 : | Isibaar | 581 | } |
130 : | edgomez | 195 | } |
131 : | Isibaar | 3 | } |
132 : | |||
133 : | edgomez | 195 | void |
134 : | MBMotionCompensation(MACROBLOCK * const mb, | ||
135 : | const uint32_t i, | ||
136 : | const uint32_t j, | ||
137 : | const IMAGE * const ref, | ||
138 : | const IMAGE * const refh, | ||
139 : | const IMAGE * const refv, | ||
140 : | const IMAGE * const refhv, | ||
141 : | IMAGE * const cur, | ||
142 : | int16_t * dct_codes, | ||
143 : | const uint32_t width, | ||
144 : | const uint32_t height, | ||
145 : | const uint32_t edged_width, | ||
146 : | Isibaar | 581 | const uint32_t quarterpel, |
147 : | suxen_drol | 701 | const int reduced_resolution, |
148 : | edgomez | 195 | const uint32_t rounding) |
149 : | Isibaar | 3 | { |
150 : | suxen_drol | 701 | |
151 : | chl | 619 | if (mb->mode == MODE_NOT_CODED || mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) { |
152 : | Isibaar | 3 | |
153 : | chl | 619 | int32_t dx = (quarterpel ? mb->qmvs[0].x : mb->mvs[0].x); |
154 : | int32_t dy = (quarterpel ? mb->qmvs[0].y : mb->mvs[0].y); | ||
155 : | chl | 530 | |
156 : | chl | 619 | if ( (mb->mode == MODE_NOT_CODED) && (dx==0) && (dy==0) ) { /* quick copy */ |
157 : | suxen_drol | 701 | transfer16x16_copy(cur->y + 16 * (i + j * edged_width), |
158 : | ref->y + 16 * (i + j * edged_width), | ||
159 : | edged_width); | ||
160 : | chl | 619 | |
161 : | transfer8x8_copy(cur->u + 8 * (i + j * edged_width/2), | ||
162 : | ref->u + 8 * (i + j * edged_width/2), | ||
163 : | edged_width / 2); | ||
164 : | transfer8x8_copy(cur->v + 8 * (i + j * edged_width/2), | ||
165 : | ref->v + 8 * (i + j * edged_width/2), | ||
166 : | edged_width / 2); | ||
167 : | return; | ||
168 : | Isibaar | 581 | } |
169 : | chl | 619 | /* quick MODE_NOT_CODED for GMC with MV!=(0,0) is still needed */ |
170 : | Isibaar | 581 | |
171 : | suxen_drol | 702 | if (reduced_resolution) |
172 : | { | ||
173 : | dx = RRV_MV_SCALEUP(dx); | ||
174 : | dy = RRV_MV_SCALEUP(dy); | ||
175 : | } | ||
176 : | |||
177 : | Isibaar | 588 | compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
178 : | suxen_drol | 701 | refv->y, refhv->y, 16 * i, 16 * j, dx, dy, |
179 : | edged_width, quarterpel, reduced_resolution, rounding); | ||
180 : | Isibaar | 3 | |
181 : | Isibaar | 581 | if (quarterpel) |
182 : | { | ||
183 : | Isibaar | 617 | dx /= 2; |
184 : | dy /= 2; | ||
185 : | Isibaar | 581 | } |
186 : | |||
187 : | Isibaar | 617 | dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
188 : | dy = (dy >> 1) + roundtab_79[dy & 0x3]; | ||
189 : | |||
190 : | suxen_drol | 702 | /* uv-block-based compensation |
191 : | XXX: rrv doesnt work with 16x16 block-based intepolation. | ||
192 : | we need to use 18x18-block interpolation */ | ||
193 : | suxen_drol | 701 | if (reduced_resolution) |
194 : | { | ||
195 : | suxen_drol | 702 | const stride = edged_width/2; |
196 : | uint8_t * current, * reference; | ||
197 : | |||
198 : | current = cur->u + 16*j*stride + 16*i; | ||
199 : | reference = refv->u + 16*j*stride + 16*i; | ||
200 : | interpolate16x16_switch(refv->u, ref->u, 16*i, 16*j, dx, dy, stride, rounding); | ||
201 : | filter_18x18_to_8x8(dct_codes + 4*64, current, stride); | ||
202 : | filter_diff_18x18_to_8x8(dct_codes + 4*64, reference, stride); | ||
203 : | transfer16x16_copy(current, reference, stride); | ||
204 : | |||
205 : | current = cur->v + 16*j*stride + 16*i; | ||
206 : | reference = refv->v + 16*j*stride + 16*i; | ||
207 : | interpolate16x16_switch(refv->v, ref->v, 16*i, 16*j, dx, dy, stride, rounding); | ||
208 : | filter_18x18_to_8x8(dct_codes + 5*64, current, stride); | ||
209 : | filter_diff_18x18_to_8x8(dct_codes + 5*64, reference, stride); | ||
210 : | transfer16x16_copy(current, reference, stride); | ||
211 : | suxen_drol | 701 | }else{ |
212 : | transfer_8to16sub(&dct_codes[4 * 64], | ||
213 : | cur->u + 8 * j * edged_width / 2 + 8 * i, | ||
214 : | interpolate8x8_switch2(refv->u, ref->u, 8 * i, 8 * j, | ||
215 : | dx, dy, edged_width / 2, rounding), | ||
216 : | edged_width / 2); | ||
217 : | Isibaar | 3 | |
218 : | suxen_drol | 701 | transfer_8to16sub(&dct_codes[5 * 64], |
219 : | cur->v + 8 * j * edged_width / 2 + 8 * i, | ||
220 : | interpolate8x8_switch2(refv->u, ref->v, 8 * i, 8 * j, | ||
221 : | dx, dy, edged_width / 2, rounding), | ||
222 : | edged_width / 2); | ||
223 : | } | ||
224 : | suxen_drol | 449 | |
225 : | chl | 530 | } else { // mode == MODE_INTER4V |
226 : | suxen_drol | 702 | int i; |
227 : | Isibaar | 3 | int32_t sum, dx, dy; |
228 : | suxen_drol | 702 | VECTOR mvs[4]; |
229 : | Isibaar | 3 | |
230 : | Isibaar | 581 | if(quarterpel) |
231 : | suxen_drol | 702 | for (i = 0; i < 4; i++) mvs[i] = mb->qmvs[i]; |
232 : | Isibaar | 581 | else |
233 : | suxen_drol | 702 | for (i = 0; i < 4; i++) mvs[i] = mb->mvs[i]; |
234 : | Isibaar | 581 | |
235 : | suxen_drol | 701 | if (reduced_resolution) |
236 : | { | ||
237 : | suxen_drol | 702 | for (i = 0; i < 4; i++) |
238 : | { | ||
239 : | mvs[i].x = RRV_MV_SCALEUP(mvs[i].x); | ||
240 : | mvs[i].y = RRV_MV_SCALEUP(mvs[i].y); | ||
241 : | } | ||
242 : | suxen_drol | 701 | } |
243 : | Isibaar | 3 | |
244 : | suxen_drol | 702 | compensate8x8_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
245 : | refv->y, refhv->y, 16 * i, 16 * j, mvs[0].x, | ||
246 : | mvs[0].y, edged_width, quarterpel, reduced_resolution, rounding); | ||
247 : | compensate8x8_interpolate(&dct_codes[1 * 64], cur->y, ref->y, refh->y, | ||
248 : | refv->y, refhv->y, 16 * i + 8, 16 * j, | ||
249 : | mvs[1].x, mvs[1].y, edged_width, quarterpel, reduced_resolution, rounding); | ||
250 : | compensate8x8_interpolate(&dct_codes[2 * 64], cur->y, ref->y, refh->y, | ||
251 : | refv->y, refhv->y, 16 * i, 16 * j + 8, | ||
252 : | mvs[2].x, mvs[2].y, edged_width, quarterpel, reduced_resolution, rounding); | ||
253 : | compensate8x8_interpolate(&dct_codes[3 * 64], cur->y, ref->y, refh->y, | ||
254 : | refv->y, refhv->y, 16 * i + 8, 16 * j + 8, | ||
255 : | mvs[3].x, mvs[3].y, edged_width, quarterpel, reduced_resolution, rounding); | ||
256 : | |||
257 : | Isibaar | 581 | if(quarterpel) |
258 : | Isibaar | 616 | sum = (mvs[0].x / 2) + (mvs[1].x / 2) + (mvs[2].x / 2) + (mvs[3].x / 2); |
259 : | else | ||
260 : | sum = mvs[0].x + mvs[1].x + mvs[2].x + mvs[3].x; | ||
261 : | Isibaar | 581 | |
262 : | Isibaar | 616 | dx = (sum >> 3) + roundtab_76[sum & 0xf]; |
263 : | Isibaar | 3 | |
264 : | Isibaar | 581 | if(quarterpel) |
265 : | Isibaar | 616 | sum = (mvs[0].y / 2) + (mvs[1].y / 2) + (mvs[2].y / 2) + (mvs[3].y / 2); |
266 : | else | ||
267 : | sum = mvs[0].y + mvs[1].y + mvs[2].y + mvs[3].y; | ||
268 : | Isibaar | 581 | |
269 : | Isibaar | 616 | dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
270 : | Isibaar | 3 | |
271 : | suxen_drol | 702 | |
272 : | /* uv-block-based compensation | ||
273 : | XXX: rrv doesnt work with 16x16 block-based intepolation. | ||
274 : | we need to use 18x18-block interpolation */ | ||
275 : | suxen_drol | 701 | if (reduced_resolution) |
276 : | { | ||
277 : | suxen_drol | 702 | const stride = edged_width/2; |
278 : | uint8_t * current, * reference; | ||
279 : | |||
280 : | current = cur->u + 16*j*stride + 16*i; | ||
281 : | reference = refv->u + 16*j*stride + 16*i; | ||
282 : | interpolate16x16_switch(refv->u, ref->u, 16*i, 16*j, dx, dy, stride, rounding); | ||
283 : | filter_18x18_to_8x8(dct_codes + 4*64, current, stride); | ||
284 : | filter_diff_18x18_to_8x8(dct_codes + 4*64, reference, stride); | ||
285 : | transfer16x16_copy(current, reference, stride); | ||
286 : | |||
287 : | current = cur->v + 16*j*stride + 16*i; | ||
288 : | reference = refv->v + 16*j*stride + 16*i; | ||
289 : | interpolate16x16_switch(refv->v, ref->v, 16*i, 16*j, dx, dy, stride, rounding); | ||
290 : | filter_18x18_to_8x8(dct_codes + 5*64, current, stride); | ||
291 : | filter_diff_18x18_to_8x8(dct_codes + 5*64, reference, stride); | ||
292 : | transfer16x16_copy(current, reference, stride); | ||
293 : | |||
294 : | suxen_drol | 701 | }else{ |
295 : | transfer_8to16sub(&dct_codes[4 * 64], | ||
296 : | cur->u + 8 * j * edged_width / 2 + 8 * i, | ||
297 : | interpolate8x8_switch2(refv->u, ref->u, 8 * i, 8 * j, | ||
298 : | dx, dy, edged_width / 2, rounding), | ||
299 : | edged_width / 2); | ||
300 : | Isibaar | 3 | |
301 : | suxen_drol | 701 | transfer_8to16sub(&dct_codes[5 * 64], |
302 : | cur->v + 8 * j * edged_width / 2 + 8 * i, | ||
303 : | interpolate8x8_switch2(refv->u, ref->v, 8 * i, 8 * j, | ||
304 : | dx, dy, edged_width / 2, rounding), | ||
305 : | edged_width / 2); | ||
306 : | } | ||
307 : | Isibaar | 3 | } |
308 : | } | ||
309 : | suxen_drol | 118 | |
310 : | |||
311 : | edgomez | 195 | void |
312 : | MBMotionCompensationBVOP(MBParam * pParam, | ||
313 : | MACROBLOCK * const mb, | ||
314 : | const uint32_t i, | ||
315 : | const uint32_t j, | ||
316 : | IMAGE * const cur, | ||
317 : | const IMAGE * const f_ref, | ||
318 : | const IMAGE * const f_refh, | ||
319 : | const IMAGE * const f_refv, | ||
320 : | const IMAGE * const f_refhv, | ||
321 : | const IMAGE * const b_ref, | ||
322 : | const IMAGE * const b_refh, | ||
323 : | const IMAGE * const b_refv, | ||
324 : | const IMAGE * const b_refhv, | ||
325 : | int16_t * dct_codes) | ||
326 : | suxen_drol | 118 | { |
327 : | static const uint32_t roundtab[16] = | ||
328 : | { 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2 }; | ||
329 : | |||
330 : | const int32_t edged_width = pParam->edged_width; | ||
331 : | int32_t dx, dy; | ||
332 : | int32_t b_dx, b_dy; | ||
333 : | chl | 374 | int k,sum; |
334 : | suxen_drol | 118 | int x = i; |
335 : | int y = j; | ||
336 : | syskin | 658 | uint32_t quarterpel = pParam->m_quarterpel; |
337 : | suxen_drol | 118 | |
338 : | edgomez | 195 | switch (mb->mode) { |
339 : | case MODE_FORWARD: | ||
340 : | suxen_drol | 118 | |
341 : | syskin | 658 | if (quarterpel) { |
342 : | dx = mb->qmvs[0].x; | ||
343 : | dy = mb->qmvs[0].y; | ||
344 : | } else { | ||
345 : | dx = mb->mvs[0].x; | ||
346 : | dy = mb->mvs[0].y; | ||
347 : | } | ||
348 : | suxen_drol | 118 | |
349 : | syskin | 658 | compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, f_ref->y, f_refh->y, |
350 : | f_refv->y, f_refhv->y, 16 * i, 16 * j, dx, | ||
351 : | suxen_drol | 701 | dy, edged_width, quarterpel, 0 /*reduced_resolution*/, 0); |
352 : | suxen_drol | 118 | |
353 : | syskin | 658 | if (quarterpel) { |
354 : | dx /= 2; | ||
355 : | dy /= 2; | ||
356 : | } | ||
357 : | suxen_drol | 118 | |
358 : | syskin | 658 | dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
359 : | dy = (dy >> 1) + roundtab_79[dy & 0x3]; | ||
360 : | suxen_drol | 118 | |
361 : | chl | 530 | /* uv-block-based compensation */ |
362 : | suxen_drol | 449 | transfer_8to16sub(&dct_codes[4 * 64], |
363 : | chl | 530 | cur->u + 8 * j * edged_width / 2 + 8 * i, |
364 : | interpolate8x8_switch2(f_refv->u, f_ref->u, 8 * i, 8 * j, | ||
365 : | dx, dy, edged_width / 2, 0), | ||
366 : | |||
367 : | suxen_drol | 449 | edged_width / 2); |
368 : | |||
369 : | transfer_8to16sub(&dct_codes[5 * 64], | ||
370 : | chl | 530 | cur->v + 8 * j * edged_width / 2 + 8 * i, |
371 : | interpolate8x8_switch2(f_refv->u, f_ref->v, 8 * i, 8 * j, | ||
372 : | dx, dy, edged_width / 2, 0), | ||
373 : | |||
374 : | suxen_drol | 449 | edged_width / 2); |
375 : | |||
376 : | suxen_drol | 118 | break; |
377 : | |||
378 : | edgomez | 195 | case MODE_BACKWARD: |
379 : | syskin | 658 | if (quarterpel) { |
380 : | b_dx = mb->b_qmvs[0].x; | ||
381 : | b_dy = mb->b_qmvs[0].y; | ||
382 : | } else { | ||
383 : | b_dx = mb->b_mvs[0].x; | ||
384 : | b_dy = mb->b_mvs[0].y; | ||
385 : | } | ||
386 : | suxen_drol | 118 | |
387 : | syskin | 658 | compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, b_ref->y, b_refh->y, |
388 : | b_refv->y, b_refhv->y, 16 * i, 16 * j, b_dx, | ||
389 : | suxen_drol | 701 | b_dy, edged_width, quarterpel, 0 /*reduced_resolution*/, 0); |
390 : | suxen_drol | 118 | |
391 : | syskin | 658 | if (quarterpel) { |
392 : | b_dx /= 2; | ||
393 : | b_dy /= 2; | ||
394 : | } | ||
395 : | suxen_drol | 118 | |
396 : | syskin | 658 | b_dx = (b_dx >> 1) + roundtab_79[b_dx & 0x3]; |
397 : | b_dy = (b_dy >> 1) + roundtab_79[b_dy & 0x3]; | ||
398 : | suxen_drol | 118 | |
399 : | |||
400 : | chl | 530 | /* uv-block-based compensation */ |
401 : | suxen_drol | 449 | transfer_8to16sub(&dct_codes[4 * 64], |
402 : | chl | 530 | cur->u + 8 * j * edged_width / 2 + 8 * i, |
403 : | interpolate8x8_switch2(f_refv->u, b_ref->u, 8 * i, 8 * j, | ||
404 : | b_dx, b_dy, edged_width / 2, 0), | ||
405 : | suxen_drol | 449 | |
406 : | chl | 530 | edged_width / 2); |
407 : | |||
408 : | suxen_drol | 449 | transfer_8to16sub(&dct_codes[5 * 64], |
409 : | chl | 530 | cur->v + 8 * j * edged_width / 2 + 8 * i, |
410 : | interpolate8x8_switch2(f_refv->u, b_ref->v, 8 * i, 8 * j, | ||
411 : | b_dx, b_dy, edged_width / 2, 0), | ||
412 : | suxen_drol | 449 | |
413 : | chl | 530 | edged_width / 2); |
414 : | |||
415 : | suxen_drol | 118 | break; |
416 : | |||
417 : | chl | 374 | case MODE_INTERPOLATE: /* _could_ use DIRECT, but would be overkill (no 4MV there) */ |
418 : | chl | 530 | case MODE_DIRECT_NO4V: |
419 : | chl | 374 | |
420 : | syskin | 658 | if (quarterpel) { |
421 : | dx = mb->qmvs[0].x; | ||
422 : | dy = mb->qmvs[0].y; | ||
423 : | b_dx = mb->b_qmvs[0].x; | ||
424 : | b_dy = mb->b_qmvs[0].y; | ||
425 : | chl | 374 | |
426 : | syskin | 658 | interpolate16x16_quarterpel((uint8_t *) f_refv->y, (uint8_t *) f_ref->y, (uint8_t *) f_refh->y, |
427 : | (uint8_t *) f_refh->y + 64, (uint8_t *) f_refhv->y, 16*i, 16*j, dx, dy, edged_width, 0); | ||
428 : | interpolate16x16_quarterpel((uint8_t *) b_refv->y, (uint8_t *) b_ref->y, (uint8_t *) b_refh->y, | ||
429 : | (uint8_t *) b_refh->y + 64, (uint8_t *) b_refhv->y, 16*i, 16*j, b_dx, b_dy, edged_width, 0); | ||
430 : | suxen_drol | 118 | |
431 : | syskin | 658 | for (k = 0; k < 4; k++) { |
432 : | transfer_8to16sub2(&dct_codes[k * 64], | ||
433 : | cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, | ||
434 : | f_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, | ||
435 : | b_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, | ||
436 : | edged_width); | ||
437 : | } | ||
438 : | b_dx /= 2; | ||
439 : | b_dy /= 2; | ||
440 : | dx /= 2; | ||
441 : | dy /= 2; | ||
442 : | |||
443 : | } else { | ||
444 : | dx = mb->mvs[0].x; | ||
445 : | dy = mb->mvs[0].y; | ||
446 : | b_dx = mb->b_mvs[0].x; | ||
447 : | b_dy = mb->b_mvs[0].y; | ||
448 : | |||
449 : | for (k = 0; k < 4; k++) { | ||
450 : | transfer_8to16sub2(&dct_codes[k * 64], | ||
451 : | cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, | ||
452 : | get_ref(f_ref->y, f_refh->y, f_refv->y, | ||
453 : | f_refhv->y, 2*i + (k&1), 2*j + (k>>1), 8, dx, dy, | ||
454 : | edged_width), | ||
455 : | get_ref(b_ref->y, b_refh->y, b_refv->y, | ||
456 : | b_refhv->y, 2*i + (k&1), 2 * j+(k>>1), 8, b_dx, b_dy, | ||
457 : | edged_width), | ||
458 : | edged_width); | ||
459 : | } | ||
460 : | |||
461 : | chl | 374 | } |
462 : | suxen_drol | 118 | |
463 : | |||
464 : | syskin | 658 | dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
465 : | dy = (dy >> 1) + roundtab_79[dy & 0x3]; | ||
466 : | suxen_drol | 118 | |
467 : | syskin | 658 | b_dx = (b_dx >> 1) + roundtab_79[b_dx & 0x3]; |
468 : | b_dy = (b_dy >> 1) + roundtab_79[b_dy & 0x3]; | ||
469 : | |||
470 : | chl | 530 | transfer_8to16sub2(&dct_codes[4 * 64], |
471 : | cur->u + (y * 8) * edged_width / 2 + (x * 8), | ||
472 : | interpolate8x8_switch2(f_refv->u, b_ref->u, 8 * i, 8 * j, | ||
473 : | b_dx, b_dy, edged_width / 2, 0), | ||
474 : | interpolate8x8_switch2(f_refv->u + 8, f_ref->u, 8 * i, 8 * j, | ||
475 : | dx, dy, edged_width / 2, 0), | ||
476 : | suxen_drol | 449 | edged_width / 2); |
477 : | |||
478 : | chl | 530 | transfer_8to16sub2(&dct_codes[5 * 64], |
479 : | cur->v + (y * 8) * edged_width / 2 + (x * 8), | ||
480 : | interpolate8x8_switch2(f_refv->u, b_ref->v, 8 * i, 8 * j, | ||
481 : | b_dx, b_dy, edged_width / 2, 0), | ||
482 : | interpolate8x8_switch2(f_refv->u + 8, f_ref->v, 8 * i, 8 * j, | ||
483 : | dx, dy, edged_width / 2, 0), | ||
484 : | suxen_drol | 449 | edged_width / 2); |
485 : | chl | 530 | |
486 : | suxen_drol | 118 | break; |
487 : | chl | 374 | |
488 : | case MODE_DIRECT: | ||
489 : | syskin | 658 | if (quarterpel) { |
490 : | for (k=0;k<4;k++) { | ||
491 : | suxen_drol | 118 | |
492 : | syskin | 658 | dx = mb->qmvs[k].x; |
493 : | dy = mb->qmvs[k].y; | ||
494 : | b_dx = mb->b_qmvs[k].x; | ||
495 : | b_dy = mb->b_qmvs[k].y; | ||
496 : | chl | 374 | |
497 : | syskin | 658 | interpolate8x8_quarterpel((uint8_t *) f_refv->y, |
498 : | (uint8_t *) f_ref->y, | ||
499 : | (uint8_t *) f_refh->y, | ||
500 : | (uint8_t *) f_refh->y + 64, | ||
501 : | (uint8_t *) f_refhv->y, | ||
502 : | 16*i + (k&1)*8, 16*j + (k>>1)*8, dx, dy, edged_width, 0); | ||
503 : | interpolate8x8_quarterpel((uint8_t *) b_refv->y, | ||
504 : | (uint8_t *) b_ref->y, | ||
505 : | (uint8_t *) b_refh->y, | ||
506 : | (uint8_t *) b_refh->y + 64, | ||
507 : | (uint8_t *) b_refhv->y, | ||
508 : | 16*i + (k&1)*8, 16*j + (k>>1)*8, b_dx, b_dy, edged_width, 0); | ||
509 : | chl | 374 | |
510 : | |||
511 : | syskin | 658 | transfer_8to16sub2(&dct_codes[k * 64], |
512 : | cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, | ||
513 : | f_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, | ||
514 : | b_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, | ||
515 : | edged_width); | ||
516 : | } | ||
517 : | sum = mb->qmvs[0].y/2 + mb->qmvs[1].y/2 + mb->qmvs[2].y/2 + mb->qmvs[3].y/2; | ||
518 : | dy = (sum >> 3) + roundtab_76[sum & 0xf]; | ||
519 : | sum = mb->qmvs[0].x/2 + mb->qmvs[1].x/2 + mb->qmvs[2].x/2 + mb->qmvs[3].x/2; | ||
520 : | dx = (sum >> 3) + roundtab_76[sum & 0xf]; | ||
521 : | chl | 374 | |
522 : | syskin | 658 | sum = mb->b_qmvs[0].y/2 + mb->b_qmvs[1].y/2 + mb->b_qmvs[2].y/2 + mb->b_qmvs[3].y/2; |
523 : | b_dy = (sum >> 3) + roundtab_76[sum & 0xf]; | ||
524 : | sum = mb->b_qmvs[0].x/2 + mb->b_qmvs[1].x/2 + mb->b_qmvs[2].x/2 + mb->b_qmvs[3].x/2; | ||
525 : | b_dx = (sum >> 3) + roundtab_76[sum & 0xf]; | ||
526 : | chl | 374 | |
527 : | syskin | 658 | } else { |
528 : | for (k=0;k<4;k++) { | ||
529 : | dx = mb->mvs[k].x; | ||
530 : | dy = mb->mvs[k].y; | ||
531 : | chl | 374 | |
532 : | syskin | 658 | b_dx = mb->b_mvs[k].x; |
533 : | b_dy = mb->b_mvs[k].y; | ||
534 : | chl | 374 | |
535 : | syskin | 658 | transfer_8to16sub2(&dct_codes[k * 64], |
536 : | cur->y + (i*16 + (k&1)*8) + (j*16 + (k>>1)*8 ) * edged_width, | ||
537 : | get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, | ||
538 : | 2*i + (k&1), 2*j + (k>>1), 8, dx, dy, | ||
539 : | edged_width), | ||
540 : | get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, | ||
541 : | 2*i + (k&1), 2*j + (k>>1), 8, b_dx, b_dy, | ||
542 : | edged_width), | ||
543 : | edged_width); | ||
544 : | } | ||
545 : | chl | 374 | |
546 : | syskin | 658 | sum = mb->mvs[0].x + mb->mvs[1].x + mb->mvs[2].x + mb->mvs[3].x; |
547 : | dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); | ||
548 : | chl | 374 | |
549 : | syskin | 658 | sum = mb->mvs[0].y + mb->mvs[1].y + mb->mvs[2].y + mb->mvs[3].y; |
550 : | dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); | ||
551 : | |||
552 : | sum = mb->b_mvs[0].x + mb->b_mvs[1].x + mb->b_mvs[2].x + mb->b_mvs[3].x; | ||
553 : | b_dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); | ||
554 : | |||
555 : | sum = mb->b_mvs[0].y + mb->b_mvs[1].y + mb->b_mvs[2].y + mb->b_mvs[3].y; | ||
556 : | b_dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); | ||
557 : | |||
558 : | } | ||
559 : | chl | 530 | transfer_8to16sub2(&dct_codes[4 * 64], |
560 : | cur->u + (y * 8) * edged_width / 2 + (x * 8), | ||
561 : | interpolate8x8_switch2(f_refv->u, b_ref->u, 8 * i, 8 * j, | ||
562 : | b_dx, b_dy, edged_width / 2, 0), | ||
563 : | interpolate8x8_switch2(f_refv->u + 8, f_ref->u, 8 * i, 8 * j, dx, dy, | ||
564 : | edged_width / 2, 0), | ||
565 : | suxen_drol | 449 | edged_width / 2); |
566 : | |||
567 : | chl | 530 | transfer_8to16sub2(&dct_codes[5 * 64], |
568 : | cur->v + (y * 8) * edged_width / 2 + (x * 8), | ||
569 : | interpolate8x8_switch2(f_refv->u, b_ref->v, 8 * i, 8 * j, | ||
570 : | b_dx, b_dy, edged_width / 2, 0), | ||
571 : | interpolate8x8_switch2(f_refv->u + 8, f_ref->v, 8 * i, 8 * j, | ||
572 : | dx, dy, edged_width / 2, 0), | ||
573 : | suxen_drol | 449 | edged_width / 2); |
574 : | chl | 530 | |
575 : | |||
576 : | suxen_drol | 118 | break; |
577 : | } | ||
578 : | } |
No admin address has been configured | ViewVC Help |
Powered by ViewVC 1.0.4 |